Lifelong Object Localization in Robotic Applications

摘要

One of the most common tasks in assistive robotics is to find some specific object in a home environment. Usually, this task is tackled by adding the objects of interest to a map of the environment as soon as the objects are detected by the vision system of the robot. However, these maps are usually static, and do not take into account the dynamic nature of a home, where anyone could move an object after the robot has seen it. In this paper, we propose a lifelong system to address this problem. The robot takes into account different possible locations for each object, and chooses the more probable one when it is required. We have designed a probability based system that stores possible locations for each object, and updates the probabilities of past locations based on newer detections.